Satellite-based Rabi rice paddy field mapping in India: a case study on Telangana state

Prashanth Reddy Putta, Fabio Dell’Acqua · 2025

Read Satellite-based Rabi rice paddy field mapping in India: a case study on Telangana state by Prashanth Reddy Putta, Fabio Dell’Acqua (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-05189.txt.

Introduction

Prashanth Reddy Putta and Fabio Dell'Acqua mapped winter rice fields across Telangana, India, from satellites in 2025. Rice area data guide food policy, yet small, scattered farms defeat broad-brush mapping. Their method keys on each district's own crop calendar, since planting dates shift by up to fifty days across the state.

District-tuned mapping reached over ninety-three percent overall accuracy, beating the usual regional approach by eight points. Results matched official statistics closely. Field sizes ran from tiny plots to near three hectares, all handled by the same frame. The approach suits other regions where farms are small and calendars vary.

Catalog record

CREATORPrashanth Reddy Putta, Fabio Dell’Acqua
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE93 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-05189

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Satellite-based Rabi rice paddy field mapping in India: a case study on Telangana state Prashanth Reddy Putta, Fabio Dell’Acqua Abstract Accurate rice area monitoring is critical for food security and agricultural policy in smallholder farming regions, yet conventional remote sensing approaches struggle with the spatiotemporal heterogeneity characteristic of fragmented agricultural landscapes. This study developed a phenology-driven classification framework that systematically adapts to local agro-ecological variations across 32 districts in Telangana, India during the 2018-19 Rabi rice season. The research reveals significant spatiotemporal diversity, with phenological timing varying by up to 50 days between districts and field sizes ranging from 0.01 to 2.94 hectares. Our district-specific calibration approach achieved 93.3% overall accuracy, an 8.0 percentage point improvement over conventional regional clustering methods, with strong validation against official government statistics (R² = 0.981) demonstrating excellent agreement between remotely sensed and ground truth data. The framework successfully mapped 732,345 hectares by adapting to agro-climatic variations, with Northern districts requiring extended land preparation phases (up to 55 days) while Southern districts showed compressed cultivation cycles. Field size analysis revealed accuracy declining 6.8 percentage points from medium to tiny fields, providing insights for operational monitoring in fragmented landscapes. These findings demonstrate that remote sensing frameworks must embrace rather than simplify landscape complexity, advancing region-specific agricultural monitoring approaches that maintain scientific rigor while serving practical policy and food security applications. Keywords: Rice mapping, phenology-based classification, Sentinel-2, Telangana, multi-temporal, district-specific calibration, smallholder agriculture, spectral indices 1 Introduction Rice cultivation stands at the nexus of global food security challenges, sustaining over half the world’s population and providing up to 76% of the caloric intake for Southeast Asian inhabitants [ 1 , 2 ] . With global population projected to reach 8.5-10.6 billion by 2050 [ 3 ] , the agricultural sector faces unprecedented pressure to increase production while managing resource constraints and climate uncertainties. Recent global rice market analyses indicate persistent volatility, with production reaching record levels of 543.3 million tonnes yet facing regional shortages and export restrictions that threaten food security [ 4 ] . The urgency of this challenge has been amplified by recent geopolitical disruptions and climate impacts. [ 3 ] demonstrate that India’s temporary export bans and climate-related production declines in Bangladesh and Sri Lanka illustrate the fragility of global rice supply chains. Such disruptions disproportionately affect import-dependent regions, particularly Sub-Saharan Africa and parts of Asia, where smallholder agricultural systems predominate and food security margins remain narrow. These smallholder systems, while contributing approximately 70% of global food production, exhibit remarkable spatial and temporal variability that challenges standardized monitoring approaches [ 5 , 6 ] . In this context, accurate and timely crop monitoring becomes essential for multiple stakeholders: farmers require information for optimal management decisions, governments need data for policy formulation and food security planning, and international organizations require monitoring capabilities for early warning systems and humanitarian response [ 7 ] . However, smallholder agricultural systems present distinctive challenges for remote sensing applications, including small field sizes, variable planting and harvesting times, diverse management practices, mixed cropping patterns, and fuzzy field boundaries [ 5 , 8 ] . These characteristics fundamentally alter the assumptions underlying many conventional crop mapping approaches. The landscape of agricultural remote sensing has undergone substantial transformation over the past decade, driven by advances in satellite sensor capabilities, computational power, and algorithmic sophistication. [ 9 ] identify four primary methodological classes in contemporary crop mapping research: spatial statistical methods, traditional machine learning approaches, phenology-based methods, and deep learning techniques. Traditional machine learning approaches, including Support Vector Machines, Random Forest, and Neural Networks, have demonstrated considerable success in crop mapping applications, achieving high accuracy when sufficient training data and computational resources are available [ 10 ] . Recent applications in crop improvement and sustainable production highlight the expanding role of machine learning across agricultural domains [ 11 ] . However, comprehensive reviews reveal that while deep learning approaches like CNN-LSTM combinations and multimodal fusion techniques show promise, they require substantial preprocessing, labeled datasets, and computational infrastructure [ 12 ] . A critical gap exists between the sophisticated algorithms developed for uniform agricultural landscapes and their adaptability to heterogeneous smallholder systems. [ 7 ] emphasize that while machine learning and deep learning methods achieve impressive accuracy in crop monitoring, most struggle with the spatial and temporal variability characteristic of diverse agricultural landscapes, limiting their applicability across different agro-ecological contexts. This adaptability gap has profound implications for understanding agricultural diversity across heterogeneous landscapes. Smallholder agricultural systems, which contribute approximately 70% of global food production, exhibit remarkable spatial and temporal variability that challenges standardized monitoring approaches [ 5 ] . The diversity in field sizes, cropping patterns, irrigation practices, and phenological timing creates a complex mosaic that requires adaptive methodological frameworks rather than one-size-fits-all solutions [ 6 , 13 ] . [ 14 ] demonstrate that successful agricultural monitoring in diverse landscapes requires approaches that can accommodate local variations while maintaining methodological consistency across broader regions. Despite substantial advances in agricultural remote sensing methodology, several critical gaps persist in understanding how monitoring approaches can adapt to landscape heterogeneity while maintaining operational viability. First, systematic evaluations of parameter calibration scale effects, comparing regional clustering versus administrative district-level optimization, remain limited in heterogeneous agricultural landscapes. Most studies either apply uniform parameters across entire study regions or acknowledge heterogeneity without developing systematic adaptation strategies. Second, while index-based methodologies offer distinct operational advantages, including computational efficiency, methodological transparency, and minimal training data requirements, systematic frameworks for adapting these approaches to diverse landscape conditions remain underexplored. Vegetation indices have maintained relevance in agricultural remote sensing due to their computational efficiency, interpretability, and reduced sensitivity to atmospheric effects [ 15 ] . Phenology-based crop mapping represents a methodologically distinct approach that exploits the temporal dynamics of crop development cycles rather than relying on static spectral signatures. [ 16 ] demonstrates the potential for phenology matching and cultivation pattern analysis using multi-source data integration, achieving robust rice mapping performance across diverse agricultural landscapes. Rice cultivation presents distinctive flooding, transplanting, and senescence phases that create identifiable spectral-temporal patterns. However, [ 17 ] provide comprehensive assessments revealing that while various algorithms can correctly identify rice fields with simple cropping patterns, different products show low consistency in fragmented rice fields, and cloud prevalence in subtropical/tropical regions poses persistent challenges. Current crop mapping literature reveals a methodological divide between algorithmic sophistication and landscape adaptability. Supervised machine learning models demonstrate high accuracy for crop identification in uniform landscapes, yet face challenges when applied across diverse agro-ecological conditions due to variations in data characteristics, phenological timing, and environmental heterogeneity. Deep learning approaches, while showing promise in controlled environments, often struggle with the transferability required for diverse landscape applications where training data may not be representative of the full range of local conditions [ 10 ] . This creates a methodological gap where the agricultural systems exhibiting the greatest landscape diversity are least well-served by standardized monitoring approaches. The challenge of systematic parameter calibration across administrative boundaries represents a particularly critical research gap. The choice between regional clustering approaches, which offer theoretical efficiency and simplified implementation, and district-specific calibration frameworks, which provide enhanced adaptation to local conditions, has profound implications for both classification accuracy and operational feasibility. This methodological

Read the complete text